A joint estimation method for torsional pendulum motion state and system parameters

By using the extended Kalman filter algorithm to jointly estimate the torsional pendulum motion state and system parameters, the problems of parameter separation and insufficient real-time performance in the prior art are solved. This achieves real-time and accurate estimation of torsional pendulum system parameters and state, improving the real-time performance and robustness of the measurement.

CN122835489BActive Publication Date: 2026-11-17CHANGCHUN INST OF OPTICS FINE MECHANICS & PHYSICS CHINESE ACAD OF SCI
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202611310465.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-08-27
Publication Date
2026-11-17
Estimated Expiration
2046-08-27

AI Technical Summary

Technical Problem

Existing torsional pendulum measurement methods separate system parameter identification and motion state estimation, cannot update parameters online, do not incorporate sensor errors into the model, and lack real-time performance, making it difficult to meet the high-precision estimation requirements of parameters under time-varying and environmental interference in precision measurement.

Method used

An extended Kalman filter algorithm is used to jointly estimate the torsional pendulum motion state and system parameters. The torsional pendulum motion state and system parameters are used as augmented states, and real-time joint estimation is performed through extended Kalman filtering. The estimated values ​​of the torsional pendulum motion state, system parameters, and estimated covariance are recursively output to suppress parameter drift and sensor error.

Benefits of technology

It achieves synchronous estimation of torsional pendulum motion state and system parameters, improves the real-time performance of the measurement process and the consistency of estimation results, enhances anti-interference capability and estimation accuracy, outputs uncertainty assessment information, and supports measurement quality evaluation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122835489B_ABST
    Figure CN122835489B_ABST
Patent Text Reader

Abstract

This invention belongs to the field of precision torsion pendulum measurement technology, and particularly relates to a method for joint estimation of torsion pendulum motion state and system parameters, comprising the following steps: S1, building a torsion pendulum measurement system; S2, establishing a principal rotational degree of freedom model and recursive estimation equations; establishing a dynamic model of the torsion pendulum system; constructing an augmented state vector from the torsion pendulum motion state and system parameters, establishing a nonlinear state transition model based on the dynamic model, and establishing an observation model; S3, performing extended Kalman filter recursive estimation; using the extended Kalman filter algorithm to predict and update the augmented state vector, recursively outputting the estimated values ​​of the torsion pendulum motion state, system parameters, and estimated covariance; S4, performing anomaly judgment and outputting results. This scheme can achieve synchronous real-time joint estimation of torsion pendulum angle, angular velocity, damping, stiffness, equilibrium angle, input torque coefficient, and sensor zero bias by constructing an augmented state vector and using extended Kalman filter recursion.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of precision torsional pendulum measurement technology, and particularly relates to a method for jointly estimating the torsional pendulum motion state and system parameters. Background Technology

[0002] Torsional pendulum systems are a typical type of low-frequency, high-sensitivity mechanical vibration system. Their basic structure typically includes a torsion wire, a pendulum body, a suspension or support structure, an angle or displacement readout unit, an excitation or loading unit, a data acquisition unit, and a data processing unit. Ideally, a torsion pendulum system can be approximated as a second-order dynamic system. The pendulum body undergoes torsional vibration about the principal rotational degree of freedom under the combined action of the torsion wire's restoring torque, damping torque, external input torque, and environmental disturbance torque.

[0003] In torsion pendulum measurements, the system's angle, angular velocity, damping, stiffness, equilibrium angle, and sensor bias are coupled with other parameters. However, in precision measurements, the system parameters of the torsion pendulum system are not absolutely stable: the stiffness of the torsion wire may drift due to temperature, stress relaxation, and assembly conditions; damping may change due to air pressure, air buoyancy, or air damping; and the angle readout device may have bias, scaling factor errors, and sampling noise. Relying solely on offline calibration or fixed parameter calculations before measurement can easily lead to estimation biases when parameter drift, sensor noise, and environmental disturbances are present, reducing the accuracy of subsequent state estimation and measurand calculations.

[0004] In the prior art, patent document with publication number "CN102478549A" discloses a "measurement method for an inverted torsional pendulum internal friction meter." This method calculates the phase angle difference between the excitation sinusoidal signal and the strain torsional vibration signal in a forced vibration mode, and performs full-spectrum fitting using the least squares method in a free vibration decay mode to obtain the logarithmic reduction in amplitude and the resonant frequency. The focus of this scheme is on internal friction measurement and decay curve fitting, but it does not perform real-time joint estimation of the augmented state composed of torsional pendulum angle, angular velocity, damping, stiffness, and sensor zero bias.

[0005] In the prior art, the patent document with publication number "CN102809473A" discloses "an automatic period measurement device for three-wire torsional pendulum vibration". This device improves the reliability of period timing through photoelectric sensors, microcontrollers, alarm and error prevention modules and data synchronization modules, and avoids the influence of mixed pendulums on counting timing. This solution focuses on solving the hardware and automatic timing problems of period measurement, but does not directly solve the problem of simultaneously estimating the state and parameters of the torsional pendulum system.

[0006] In the prior art, patent document with publication number "CN111537121A" discloses an "online testing method and system for sinusoidal torque device system parameters". This method collects dynamic angle, angular velocity, angular acceleration and standard torque sensor outputs by giving sinusoidal excitations of different frequencies and amplitudes, and solves for system parameters such as moment of inertia, damping and elastic coefficient. This scheme is similar to the present invention, both focusing on the identification of torsional system parameters. However, this scheme mainly relies on solving system parameters by solving simultaneous equations under multiple excitations. The motion state estimation and system parameter identification are still relatively separated, and there is insufficient recursive processing for sensor zero bias, slow parameter drift, multi-sensor noise and estimation uncertainty.

[0007] In the prior art, patent document with publication number "CN113375863A" discloses "a method for determining the moment of inertia of a rigid body based on the torsion pendulum method". This method obtains the inertial measurement action parameters of the torsion pendulum measurement system, the mass of the rigid body, the distance from the center of gravity to the torsion pendulum axis, the torsion pendulum period, the damping coefficient, and the aerodynamic damping coefficient, etc., calculates the moment of inertia of the rigid body and considers aerodynamic damping correction. This scheme focuses on the calculation of moment of inertia and damping correction, but does not provide a recursive estimation method that updates the motion state and multiple system parameters simultaneously during the measurement process.

[0008] In the prior art, the patent document with publication number "CN118244377A" discloses "a gravimeter calibration method tracing back to the law of universal gravitation". This method reflects the importance of period, angular acceleration and uncertainty assessment in high-precision torsion balance or torsion pendulum measurement. However, its core is gravimeter calibration and gravitational source tracing, rather than joint estimation of state and parameters for general torsion pendulum systems.

[0009] In the prior art, the patent document with publication number "CN110500034A" discloses "a method for establishing a neural network model, determining the parameters of the torsion pendulum drill string and directional drilling". It discloses a method for training a neural network model through historical operation data to determine the control parameters of the torsion pendulum drill string during sliding directional drilling. This method is different from the precision torsion pendulum measurement system targeted by this invention. At the same time, its data-driven model does not directly output the physical parameters of the torsion pendulum system and its estimation uncertainty.

[0010] In summary, existing methods for measuring torsional pendulums mainly have the following problems: 1. Existing technologies typically handle system parameter identification and motion state estimation separately. A common approach is to first obtain parameters such as damping, stiffness, and moment of inertia through period measurement, peak decay fitting, sine fitting, or multiple sets of simultaneous equations, and then use these parameters as fixed values ​​for subsequent calculations. This approach does not fully consider the slow drift of parameters over time, nor does it incorporate sensor bias, scaling factor error, and measurement noise into the torsional dynamics model.

[0011] 2. In the actual operation of the torsion pendulum system, the angle readout signal is affected by noise and zero-point drift. The stiffness and damping of the torsion wire are affected by factors such as temperature, air pressure, material relaxation, and installation status. External disturbances can also cause the torsion pendulum motion to deviate from the ideal model. Fixed parameter calculations will lead to inaccurate estimations of angle, angular velocity, angular acceleration, and measured parameters.

[0012] 3. Fitting methods based on complete decay curves or multiple sets of steady-state sinusoidal excitations can only calculate results after a measurement period has ended, which is not real-time enough; and some methods can only output the final parameter values, but cannot output the covariance or reliability of the estimated results at each time step, which is not convenient for automatic control and measurement quality judgment.

[0013] As can be seen from the above, existing torsional pendulum measurement methods generally suffer from problems such as separate processing of state and parameters, inability to update parameters online, failure to incorporate sensor errors into the model, and insufficient real-time performance. These problems make it difficult to meet the high-precision estimation requirements of parameters under time-varying and environmental interference conditions in precision measurement. Therefore, it is urgent to propose a method that can solve the above problems. Summary of the Invention

[0014] In view of this, the present invention aims to provide a method for joint estimation of torsional pendulum motion state and system parameters. The torsional pendulum motion state and system parameters are used as augmented states and real-time joint estimation is performed based on extended Kalman filtering to solve the problems of mutual coupling between motion state and system parameters, parameter drift and sensor error leading to decreased estimation accuracy during torsional pendulum measurement.

[0015] To achieve the above objectives, the technical solution created by this invention is implemented as follows: A method for jointly estimating torsional pendulum motion state and system parameters includes the following steps: S1. Construct a torsional pendulum measurement system; S2. Establish the main rotational degree of freedom model and recursive estimation equations; take the angle around the torsion axis as the main rotational degree of freedom, establish the dynamic model of the torsion pendulum system; construct the augmented state vector by combining the torsion pendulum motion state with the system parameters, establish a nonlinear state transition model based on the dynamic model, and establish an observation model. S3. Perform extended Kalman filter recursive estimation; use the extended Kalman filter algorithm to predict and update the augmented state vector, and recursively output the estimated values ​​of the torsional motion state, system parameters, and estimated covariance. S4. Perform anomaly detection and output results: Determine whether the measured value is abnormal based on the normalized innovation square. If the measured value is abnormal, reduce the weight of the current observation or skip the observation update at the current time. If the measured value is normal, continue the recursive estimation and output the estimated value of the torsional motion state, the estimated value of the system parameters, and the estimated covariance.

[0016] Furthermore, the torsion pendulum measurement system in S1 includes: a torsion pendulum mechanical unit, an excitation or loading unit, a readout unit, a data acquisition unit, and a data processing unit; the torsion pendulum mechanical unit includes a torsion wire, a pendulum body, and a support structure; the excitation or loading unit is a device for applying a known or controllable input torque to the pendulum body; the readout unit is a sensor for measuring the torsion angle, displacement, angular velocity, or torque of the pendulum body; and the data processing unit is used to execute the extended Kalman filter algorithm. Temperature, air pressure, and vibration environment monitoring are enabled during torsion measurement. For input signals, sensor signals, and environmental signals with different sampling rates, timestamps are used for synchronization.

[0017] Furthermore, the dynamic model of the torsional yaw system in S2 is expressed by the following equation: ;in, I This represents the equivalent moment of inertia of the torsional yaw system. c Indicates the equivalent damping coefficient; Indicates the linear torsional stiffness coefficient; Indicates the angle of torsion; Indicates the angular velocity of the torsional yaw; Indicates angular acceleration; Indicates the equilibrium angle; Indicates the input signal; Indicates the input torque coefficient; This indicates an unmodeled disturbance torque; The dynamic model of the torsional pendulum system is expressed in normalized form as follows: ; ;in, Indicates the angular velocity of the torsional yaw; Indicates angular acceleration; Represents the equivalent damping parameter. ; Represents the equivalent stiffness parameter. ; Indicates the normalized input parameters. .

[0018] Furthermore, the augmented state vector in S2 is: ;in, Indicates the first k The augmented state vector at each sampling time; Indicates the first k The torsional angle at each sampling moment; Indicates the first k The angular velocity of the torsion at each sampling moment; Indicates the first k The equivalent damping parameters at each sampling time; Indicates the first k Equivalent stiffness parameters at each sampling time; Indicates the first k The balance angle at each sampling time; Indicates the first k Normalized input parameters at each sampling time; Indicates the first k The angle at each sampling time is read out as zero bias; Indicates the first k The scaling factor is read from the angle at each sampling time. express The transpose of .

[0019] Furthermore, the augmented state vector in S2 is: ;in, Indicates the first k The augmented state vector at each sampling time; Indicates the first k The torsional angle at each sampling moment; Indicates the first k The angular velocity of the torsion at each sampling moment; Indicates the first k The equivalent moment of inertia at each sampling time; Indicates the first k The equivalent damping coefficient at each sampling time; This represents the torsional stiffness coefficient at the k-th sampling time. Indicates the first k The balance angle at each sampling time; Indicates the first k Input torque coefficient at each sampling time; This indicates that the angle readout at the k-th sampling time is zero bias; express The transpose of .

[0020] Furthermore, the system parameters in the augmented state vector change with time, and its state transition is expressed as: ;in, Indicates the first k The system parameters to be estimated at +1 sampling time; Indicates the first k The system parameters to be estimated at each sampling time, Indicates the first k The parameter process noise at each sampling time; by setting The covariance control algorithm tracks parameter drift speed; The observation model is: ;in, Indicates the first k Angle sensor measurement values ​​at each sampling time; Indicates the first k The scaling factor is read from the angle at each sampling time. Indicates the first k The torsional angle at each sampling moment; Indicates the first k The angle at each sampling time is read out as zero bias; This represents the angle measurement noise at the k-th sampling time.

[0021] Furthermore, the extended Kalman filter recursive estimation includes: The state prediction is based on the dynamic model of the torsional pendulum system and the state estimate from the previous moment, expressed by the following formula: in, Indicates the first k The predicted state value at each sampling time; Indicates the first k -1 state estimate at sampling time; Indicates the first k -1 sampling time of the input signal; This represents the nonlinear state transition function obtained by discretizing the dynamic model of the torsion pendulum system. Indicates the sampling period; The state transition Jacobian matrix and covariance matrix are calculated and predicted by the following formula: ; ;in, This represents the state transition Jacobian matrix at the k-th sampling time. express The transpose of the matrix; Indicates the first k -1 state estimate at sampling time; Indicates the first k State prediction covariance matrix at each sampling time; Indicates the first k -1 sampling time point state estimation covariance matrix; This represents the process noise covariance matrix at the k-th sampling time. Represents the nonlinear state transition function For the state vector The partial derivatives; The predicted observations, the observation Jacobian matrix, and the innovation quantity are calculated based on the observation model and are expressed by the following formula: ; ; ;in, This represents the predicted observation value at the k-th sampling time. Represents the observation function; This represents the predicted state value at the k-th sampling time. Represents the observation Jacobian matrix at the k-th sampling time; This represents the amount of innovation at the k-th sampling time. This represents the observation value at the k-th sampling time. Represents the observation function For the state vector The partial derivatives; The innovation covariance matrix and Kalman gain are calculated using the following formula: ; ;in, This represents the innovation covariance matrix at the k-th sampling time. Represents the observation Jacobian matrix at the k-th sampling time; express The transpose of the matrix; Indicates the first k State prediction covariance matrix at each sampling time; This represents the measurement noise covariance matrix at the k-th sampling time. This represents the Kalman gain at the k-th sampling time. express The inverse matrix; The updated augmented state vector and covariance matrix are expressed by the following equation: ; ;in, Indicates the first k State estimates at each sampling time; Indicates the first k The predicted state value at each sampling time; This represents the Kalman gain at the k-th sampling time. express The transpose of the matrix; This represents the amount of innovation at the k-th sampling time. This represents the estimation of the covariance matrix; Represents the identity matrix; Represents the observation Jacobian matrix at the k-th sampling time; express The transpose of the matrix; Indicates the first kState prediction covariance matrix at each sampling time; Let represent the measurement noise covariance matrix at the k-th sampling time.

[0022] Furthermore, the normalized squared innovation in S4 is calculated as follows: ;in, Represents the normalized squared innovation at the k-th sampling time; express The transpose of the matrix; express The inverse matrix; where, This represents the innovation covariance matrix at the k-th sampling time. This represents the amount of innovation at the k-th sampling time. when If the value exceeds a preset threshold, the current observation is determined to be an abnormal measurement value, and the corresponding measurement noise covariance is increased, the weight of the current observation is reduced, or the observation update at the current time is skipped. When the changes in motion state and system parameters in the augmented state vector are less than a set threshold within M consecutive sampling windows, and the average value of innovation is close to zero and the magnitude of innovation covariance matrix is ​​within a set range, the estimation of the augmented state vector is determined to have entered a stable phase; where M is the sliding window length.

[0023] Furthermore, the readout unit includes a multi-channel capacitive displacement sensor, which includes at least two channels arranged along the X direction: the voltage signal collected in real time by the multi-channel capacitive displacement sensor is converted into a torsional angle observation, and the torsional angle observation is input into an extended Kalman filter for recursive estimation point by point; For the i A capacitive displacement sensor converts voltage into displacement, expressed by the following formula. ;in, This represents the displacement measured by the i-th capacitive displacement sensor; This represents the voltage measurement value of the i-th capacitive displacement sensor; This indicates the zero-position bias of the i-th capacitive displacement sensor; This represents the proportional coefficient of the i-th capacitive displacement sensor.

[0024] Furthermore, the multi-channel capacitive displacement sensor includes two channels arranged along the X-axis; for the two capacitive displacement sensors in the X-axis, the yaw angle around the z-axis is calculated according to the following relationship: ;in, Indicates the yaw angle about the z-axis; , These represent the displacements measured by the two capacitive displacement sensors, respectively. This indicates the spacing parameter between the two capacitive displacement sensors.

[0025] Compared with the prior art, the present invention can achieve the following beneficial effects: 1. Synchronous joint estimation of state and parameters to improve real-time performance and consistency: This invention constructs an augmented state vector from the torsional motion state (angle, angular velocity) and system parameters (damping, stiffness, equilibrium angle, input torque coefficient, sensor zero bias), and completes recursive estimation synchronously in the same extended Kalman filter framework. There is no need to separate parameter identification and state estimation into two independent steps, which effectively solves the problem of mutual coupling between the two and improves the real-time performance of the measurement process and the consistency of the estimation results.

[0026] 2. Suppressing parameter drift and sensor error, enhancing robustness: This invention effectively addresses practical engineering problems such as the influence of temperature on the stiffness of the torsion wire, the damping effect of air pressure changes, and the presence of zero bias and scaling factor errors in the sensor by updating system parameters online. At the same time, it incorporates sensor zero bias, scaling factor error, and measurement noise into the observation model. Compared with traditional periodic measurement, peak method, and simultaneous equation solving methods, it has stronger anti-interference ability and estimation accuracy in measurement environments with slow parameter changes and sensor errors.

[0027] 3. Output uncertainty assessment information to support measurement quality evaluation: During the recursive estimation process, this invention can simultaneously output the estimated covariance matrix, the sequence of innovative quantities, and the results of outlier judgment, providing a quantitative basis for the confidence assessment of measurement results, data quality monitoring, and abnormal data identification, thus making up for the shortcomings of traditional methods in quantifying the estimation uncertainty.

[0028] 4. Good compatibility, strong real-time performance, and easy engineering deployment: This invention is mainly implemented through data processing algorithms, without the need for hardware modification of existing torsional pendulum measurement devices, resulting in low implementation costs; it can directly access multi-channel sensor data collected in practice, and complete the entire process of observation calculation, state prediction, observation update and parameter recursion within each sampling period, which can meet the real-time processing requirements in the torsional pendulum measurement process. Attached Figure Description

[0029] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is an overall flowchart of the method for jointly estimating the torsional pendulum motion state and system parameters in an embodiment of the present invention; Figure 2 This is a schematic diagram of the twisted wire, pendulum body, and support structure in an embodiment of the present invention; Figure 3This is a schematic diagram of the torsional oscillation dynamics model of the principal rotational degree of freedom about the z-axis in an embodiment of the present invention; Figure 4 This is a graph showing the measured voltage signal curves of the six-channel capacitive sensors in an embodiment of the present invention.

[0030] Figure reference numerals: 1. Twisted wire; 2. Pendulum body; 3. Support structure. Detailed Implementation

[0031] To make the purpose, technical solution, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1-4 The present invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and do not constitute a limitation thereof.

[0032] This invention provides a method for jointly estimating the torsional oscillation state and system parameters, comprising the following steps: S1. Construct a torsion pendulum measurement system; the torsion pendulum measurement system includes: a torsion pendulum mechanical unit, an excitation or loading unit, a readout unit, a data acquisition unit, and a data processing unit.

[0033] The torsion pendulum mechanical unit includes a torsion wire 1, a pendulum body 2, and a support structure 3. The excitation or loading unit is a device for applying a known or controllable input torque to the pendulum body 2. The excitation or loading unit includes an electromagnetic torque generator, an electrostatic torque generator, a turntable, standard weights, a standard torque sensor, or other devices capable of generating a known input. The readout unit is a sensor for measuring the torsion angle, displacement, angular velocity, or torque of the pendulum body 2. The readout unit includes a grating, a photoelectric sensor, an autocollimator, a capacitive displacement sensor, a laser interferometer, an angle encoder, or a torque sensor. The data processing unit executes an extended Kalman filter algorithm. The electromagnetic torque generator applies a controllable input torque to the pendulum body. The data acquisition unit simultaneously acquires the angle readout signal, the electromagnetic torque generator input signal, and the temperature signal.

[0034] Temperature, air pressure, and vibration environment monitoring are enabled during high-precision torsion pendulum measurement, and the torsion pendulum system is made to work in a stable environment. For input signals, sensor signals, and environmental signals with different sampling rates, timestamps are used for synchronization.

[0035] S2. Establish the main rotational degree of freedom model and recursive estimation equations; take the angle around the torsion axis as the main rotational degree of freedom to establish the dynamic model of the torsion pendulum system; construct an augmented state vector by combining the torsion pendulum motion state with the system parameters, establish a nonlinear state transition model based on the dynamic model, and establish an observation model.

[0036] The dynamic model of the torsional pendulum system is represented by the following equation: ;in, IThis represents the equivalent moment of inertia of the torsional yaw system. c Indicates the equivalent damping coefficient; Indicates the linear torsional stiffness coefficient; Indicates the angle of torsion; Indicates the angular velocity of the torsional yaw; Indicates angular acceleration; Indicates the equilibrium angle; Indicates the input signal; Indicates the input torque coefficient; This indicates an unmodeled disturbance torque; In some applications, without an absolutely known input torque, it is difficult to simultaneously identify I, c, and The absolute value of is given, and the dynamic model of the torsional pendulum system is expressed in normalized form as follows: ; ;in, Indicates the angular velocity of the torsional yaw; Indicates angular acceleration; Represents the equivalent damping parameter. ; Represents the equivalent stiffness parameter. ; Indicates the normalized input parameters. Unmodeled disturbance torque In filtering algorithms, the process noise covariance can be used to characterize it.

[0037] The data processing unit establishes an augmented state vector, simultaneously writing the motion states such as torsional angle and angular velocity, as well as system parameters, into the state vector. The augmented state vector is: ;in, Indicates the first k The augmented state vector at each sampling time; Indicates the first k The torsional angle at each sampling moment; Indicates the first k The angular velocity of the torsion at each sampling moment; Indicates the first k The equivalent damping parameters at each sampling time; Indicates the first k Equivalent stiffness parameters at each sampling time; Indicates the first k The balance angle at each sampling time; Indicates the first k Normalized input parameters at each sampling time; Indicates the first k The angle at each sampling time is read out as zero bias; Indicates the firstk The scaling factor is read from the angle at each sampling time. express The transpose of .

[0038] Given a standard torque sensor, known electromagnetic torque, known electrostatic torque, or known change in moment of inertia, the augmented state vector can be written as: ;in, Indicates the first k The augmented state vector at each sampling time; Indicates the first k The torsional angle at each sampling moment; Indicates the first k The angular velocity of the torsion at each sampling moment; Indicates the first k The equivalent moment of inertia at each sampling time; Indicates the first k The equivalent damping coefficient at each sampling time; This represents the torsional stiffness coefficient at the k-th sampling time. Indicates the first k The balance angle at each sampling time; Indicates the first k Input torque coefficient at each sampling time; This indicates that the angle readout at the k-th sampling time is zero bias; express The transpose of .

[0039] At this point, the moment of inertia can be directly estimated. The equivalent damping coefficient at the k-th sampling time Torsional stiffness coefficient at the kth sampling time The input torque coefficient at the kth sampling time To ensure that the physical parameters are positive, it is also possible to... , , Estimation is performed after taking the logarithm.

[0040] The system parameters in the augmented state vector change slowly over time, and its state transition is represented as: ;in, Indicates the first k The system parameters to be estimated at +1 sampling time; Indicates the first k The system parameters to be estimated at each sampling time, Indicates the first k The parameter process noise at each sampling time; by setting The covariance control algorithm tracks parameter drift speed.

[0041] The data processing unit establishes an observation model. Taking angle readout as an example, the observation model for sensor measurements is expressed as follows: ;in, Indicates the first k Angle sensor measurement values ​​at each sampling time; Indicates the first k The scaling factor is read from the angle at each sampling time. Indicates the first k The torsional angle at each sampling moment; Indicates the first k The angle at each sampling time is read out as zero bias; This represents the angle measurement noise at the k-th sampling time.

[0042] If the readout unit measures the edge displacement of pendulum 2, the length of the pendulum arm is... B Then the displacement observation model can be expressed as: ;in, Indicates the first k Displacement sensor measurements at each sampling time; B Indicates the length of the swing arm; Indicates the first k The torsional angle at each sampling moment; This indicates that the displacement readout at the k-th sampling time is zero bias. This represents the displacement measurement noise at the k-th sampling time.

[0043] At small angles It can be approximated as: .

[0044] In multi-sensor measurement scenarios, the measurements from angle sensors, displacement sensors, angular velocity sensors, torque sensors, and environmental sensors can be combined to form an observation vector, which is expressed by the following formula: ;in, Indicates the first k The observation vector at each sampling time; Indicates the first k Angle sensor measurement values ​​at each sampling time; Indicates the first k Displacement sensor measurements at each sampling time; Indicates the first k Angular velocity sensor measurements at each sampling time; Indicates the first kTorque sensor measurements at each sampling time; Indicates the first k Environmental sensor measurements at each sampling time; express The transpose of .

[0045] Different sensors can have different sampling frequencies. When only some sensor data is available at a certain moment, only the corresponding observation equation is used for updating.

[0046] S3. Perform extended Kalman filter recursive estimation; use the extended Kalman filter algorithm to predict and update the augmented state vector, and recursively output the estimated values ​​of the torsional motion state, the estimated values ​​of the system parameters, and the estimated covariance.

[0047] The data processing unit uses the Extended Kalman Filter (EKF) algorithm to recursively estimate the augmented state vector. The EKF recursive estimation includes: The state prediction is based on the dynamic model of the torsional pendulum system and the state estimate from the previous moment, expressed by the following formula: in, Indicates the first k The predicted state value at each sampling time; Indicates the first k -1 state estimate at sampling time; Indicates the first k -1 sampling time of the input signal; This represents the nonlinear state transition function obtained by discretizing the dynamic model of the torsion pendulum system. The sampling period is indicated; the discretization method can be Euler's method, fourth-order Runge-Kutta method, or other numerical integration methods.

[0048] The state transition Jacobian matrix and covariance matrix are calculated and predicted by the following formula: ; ;in, This represents the state transition Jacobian matrix at the k-th sampling time. express The transpose of the matrix; Indicates the first k -1 state estimate at sampling time; Indicates the first k State prediction covariance matrix at each sampling time; Indicates the first k -1 sampling time point state estimation covariance matrix; This represents the process noise covariance matrix at the k-th sampling time. Represents the nonlinear state transition function For the state vector The partial derivatives of .

[0049] The predicted observations, the observation Jacobian matrix, and the innovation quantity are calculated based on the observation model and are expressed by the following formula: ; ; ;in, This represents the predicted observation value at the k-th sampling time. Represents the observation function; This represents the predicted state value at the k-th sampling time. Represents the observation Jacobian matrix at the k-th sampling time; This represents the amount of innovation at the k-th sampling time. This represents the observation value at the k-th sampling time. Represents the observation function For the state vector The partial derivatives of .

[0050] The innovation covariance matrix and Kalman gain are calculated using the following formula: ; ;in, This represents the innovation covariance matrix at the k-th sampling time. Represents the observation Jacobian matrix at the k-th sampling time; express The transpose of the matrix; Indicates the first k State prediction covariance matrix at each sampling time; This represents the measurement noise covariance matrix at the k-th sampling time. This represents the Kalman gain at the k-th sampling time. express The inverse matrix.

[0051] Updating the augmented state vector and covariance matrix can further improve numerical stability, as expressed by the following equation: ; ;in, Indicates the first k State estimates at each sampling time; Indicates the first k The predicted state value at each sampling time; This represents the Kalman gain at the k-th sampling time. express The transpose of the matrix; This represents the amount of innovation at the k-th sampling time. This represents the estimation of the covariance matrix; Represents the identity matrix; Represents the observation Jacobian matrix at the k-th sampling time; express The transpose of the matrix; Indicates the first k State prediction covariance matrix at each sampling time; Let represent the measurement noise covariance matrix at the k-th sampling time.

[0052] S4. Perform anomaly detection and output results: Determine whether the measured value is abnormal based on the normalized innovation square. If the measured value is abnormal, reduce the weight of the current observation or skip the observation update at the current time. If the measured value is normal, continue the recursive estimation and output the estimated value of the torsional motion state, the estimated value of the system parameters, and the estimated covariance.

[0053] To determine measurement anomalies and parameter convergence status, and considering the influence of sensor noise and external disturbances, the normalized squared innovation is calculated, expressed by the following formula: ;in, Indicates the first k Normalized innovation square at each sampling time; express The transpose of the matrix; express The inverse matrix; where, Indicates the first k The innovative covariance matrix at each sampling time; Indicates the first k The amount of innovation at each sampling moment; when If the value exceeds a preset threshold, the current observation is determined to be an abnormal measurement value. The corresponding measurement noise covariance is increased, the weight of the current observation is reduced, or the observation update at the current time is skipped.

[0054] When the changes in motion state and system parameters in the augmented state vector are less than a set threshold within M consecutive sampling windows, and the average value of innovation is close to zero and the magnitude of innovation covariance matrix is ​​within a set range, the estimation of the augmented state vector is determined to have entered a stable phase; where M is the sliding window length.

[0055] The estimation of the augmented state vector has entered a stable phase, as expressed by the following equation: ;in, Indicates the first k - iSystem parameter estimates at each sampling time; Indicates the first k - i- System parameter estimates at one sampling time; This represents the parameter convergence threshold, which can be set according to the specific device and measurement noise; M is the sliding window length.

[0056] The readout unit includes a multi-channel capacitive displacement sensor, which includes at least two channels arranged along the X direction: the voltage signal collected in real time by the multi-channel capacitive displacement sensor is converted into a torsional angle observation, and the torsional angle observation is input into the extended Kalman filter for recursive estimation point by point.

[0057] In one embodiment, the multi-channel capacitive displacement sensor includes two channels arranged along the X direction, namely X1 and X2. The observed measurement is obtained by calculating the principal rotational degree of freedom angle around the z-axis from the measured voltages of the two capacitive displacement sensors X1 and X2. When only some sensor data is available at a certain sampling time, only the corresponding observation equation is used for extended Kalman filtering update.

[0058] The data processing unit converts the voltages X1 and X2 into corresponding displacements based on the calibration relationship of the capacitive displacement sensor. For the first... i A capacitive displacement sensor converts voltage into displacement. The sensor voltage, proportional coefficient, and zero-position offset are all determined by the sensor calibration results, specifically expressed by the following formula: ;in, Indicates the first i Displacement measured by a capacitive displacement sensor; Indicates the first i Voltage measurement value of the road capacitive displacement sensor; Indicates the first i Zero-position bias of the capacitive displacement sensor; Indicates the first i The proportional coefficient of the road capacitive displacement sensor.

[0059] For a dual-channel capacitive displacement sensor in the X direction, the yaw angle around the z-axis is calculated according to the following relationship: ;in, Indicates the yaw angle about the z-axis; , These represent the displacements measured by the two capacitive displacement sensors, respectively. This indicates the spacing parameter between the two capacitive displacement sensors.

[0060] In this embodiment, the torsional yaw angle around the z-axis is selected as the angular observation of the main rotational degree of freedom, and the sampled points are input into the extended Kalman filter.

[0061] The extended Kalman filter recursive estimation is performed. The angle, angular velocity and system parameters at the current time are predicted using the state estimate of the previous time step, the input signal and the torsional dynamic model. The innovation is then calculated using the current measurement value. The motion state and system parameters are corrected simultaneously according to the Kalman gain. Finally, the covariance matrix is ​​updated to obtain the uncertainty of the estimation result at the current time step.

[0062] In practical applications, the data processing unit completes a prediction and update every time it receives data at a sampling time. First, it reads the sensor data at the current sampling time and forms an observation. Then, it uses the state estimate and sampling period from the previous sampling time to calculate the predicted state. Next, it calculates the innovation quantity, innovation covariance, and Kalman gain from the current observation. Finally, it synchronously corrects the torsional angle, angular velocity, equivalent damping parameter, equivalent stiffness parameter, equilibrium angle, and sensor zero bias, and outputs the estimated value and covariance at the current sampling time.

[0063] In one implementation, the following discrete state prediction equation and observation equation are used: ; ; ; ; in, Indicates the first k +1 sampling time of the torsion angle; Indicates the first k The torsional angle at each sampling moment; Indicates the sampling period; Indicates the first k The angular velocity of the torsion at each sampling moment; Indicates the first k +1 sampling time point to the torsional angular velocity; Indicates the first k The equivalent damping parameters at each sampling time; Indicates the first k +1 equivalent damping parameters at sampling time; Indicates the first k Equivalent stiffness parameters at each sampling time; Indicates the first k +1 equivalent stiffness parameter at sampling time; Indicates the first k The balance angle at each sampling time; Indicates the first k The balance angle at +1 sampling time; Indicates the first k The angle at each sampling time is read out as zero bias; Indicates the first kAngle sensor measurement values ​​at each sampling time; Indicates the first k The scaling factor is read from the angle at each sampling time. This represents the angle measurement noise at the k-th sampling time.

[0064] Among them, the observed measurement consists of the pendulum angle calculated from the measured data of the capacitive displacement sensor, the angle readout scaling factor is used to characterize the sensor readout proportional error, the angle readout zero bias is used to characterize the sensor zero-point drift, and the angle measurement noise is used to characterize the sensor random error.

[0065] Figure 1 The flowchart shows the overall process of the joint estimation method for torsional pendulum motion state and system parameters. Figure 2 The diagram shows the twisted wire 1, the pendulum body 2, and the support structure 3, as follows. Figure 2 As shown, the upper end of the twisted wire 1 is fixed to the support structure 3, and the lower end is connected to the pendulum body 2.

[0066] Figure 3 This is a schematic diagram of the dynamic model of a torsional pendulum with a principal rotational degree of freedom about the z-axis. The pendulum 2 twists about the z-axis of the torsion wire 1. This twisting is caused by the combined effects of the restoring torque, damping torque, electromagnetic input torque, and external disturbance torque of the torsion wire 1. The restoring torque of the torsion wire 1 is related to the angular displacement of the pendulum 2 relative to the equilibrium angle, the damping torque is related to the angular velocity, and the electromagnetic input torque is related to the input signal. Figure 3 In the diagram, z represents the axial direction of twisted wire 1, a represents the restoring torque, b represents the damping torque, c represents the electromagnetic input torque, d represents the external disturbance torque, and e represents the torsion angle around the z-axis.

[0067] In one embodiment, the readout unit includes six capacitive displacement sensors, specifically X1, X2, Y1, Y2, Z1, and Z2. X1 and X2 are two channels arranged along the X-axis, used for differential calculation of the principal rotational degree of freedom angles around the z-axis. Y1, Y2, Z1, and Z2 are used to measure the motion of other degrees of freedom and for environmental compensation. Specifically, the data processing unit first converts the voltage of each channel into displacement using a formula, then uses the differential displacement of X1 and X2 to calculate the torsional angle around the z-axis, and uses this as the observation of the principal rotational degree of freedom. The sampling frequency is 10 Hz, and the measured six-channel voltage signals are as follows: Figure 4 As shown, Figure 4 In the diagram, the horizontal axis represents time in hours (h), and the vertical axis represents sensor voltage in volts (v). The blue line at L1 represents the voltage of X1, the green line at L2 represents the voltage of Y1, the purple line at L3 represents the voltage of Z1, the orange line at L4 represents the voltage of X2, the red line at L5 represents the voltage of Y2, and the brown line at L6 represents the voltage of Z2.

[0068] It should be noted that the present invention is not limited to the filtering algorithm, sensor type and excitation method specifically used in the above embodiments. In specific applications, the filtering algorithm can adopt a variety of nonlinear estimation algorithms, including but not limited to extended Kalman filtering, unscented Kalman filtering, comprehension Kalman filtering, ensemble Kalman filtering, particle filtering, square root Kalman filtering, etc.

[0069] The system parameters in the augmented state vector can be adjusted according to the specific device. For a torsion pendulum system without an absolute torque input, the equivalent damping parameters and equivalent stiffness parameters can be estimated. For a torsion pendulum system with a known calibrated torque, a standard torque sensor, or a known change in moment of inertia, the moment of inertia, damping coefficient, and torsional stiffness coefficient can be directly estimated.

[0070] For torsional gyroscope systems with significant nonlinear restoring torque or nonlinear damping, nonlinear stiffness, nonlinear damping, or friction terms can be added to the state transition model, and the corresponding coefficients can be estimated as part of the augmented state.

[0071] Observation sensors can take many forms, including but not limited to gratings, autocollimators, photoelectric sensors, capacitive displacement sensors, laser interferometers, angle encoders, torque sensors, accelerometers, etc. Different sensors correspond to different observation models, but their basic idea is to use the observation model to link the sensor measurement values ​​with the augmented state vector.

[0072] The input excitation method can take many forms, including free decay release, single-frequency sine excitation, multi-frequency sine excitation, frequency sweep excitation, step excitation, pseudo-random excitation or natural perturbation. Different excitation methods will affect the observability of parameters, but do not change the core idea of ​​this invention to jointly estimate the state and parameters by augmenting the state and filtering algorithm.

[0073] Considering that the torsional pendulum measurement signal is affected by noise, various outlier handling and filtering strategies can be adopted, including innovation threshold judgment, normalized innovation square test, moving mean filtering, median filtering, robust weight adjustment, and adaptive adjustment of measurement noise covariance.

[0074] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0075] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for estimating the state of a torsional pendulum motion and system parameters, characterized in that, Includes the following steps: S1. Construct a torsional pendulum measurement system; S2. Establish the main rotational degree of freedom model and recursive estimation equations; take the angle around the torsion axis as the main rotational degree of freedom, establish the dynamic model of the torsion pendulum system; construct the augmented state vector by combining the torsion pendulum motion state with the system parameters, establish a nonlinear state transition model based on the dynamic model, and establish an observation model. The augmented state vector is: ;in, Indicates the first k The augmented state vector at each sampling time; Indicates the first k The torsional angle at each sampling moment; Indicates the first k The angular velocity of the torsion at each sampling moment; Indicates the first k The equivalent damping parameters at each sampling time; Indicates the first k Equivalent stiffness parameters at each sampling time; Indicates the first k The balance angle at each sampling time; Indicates the first k Normalized input parameters at each sampling time; Indicates the first k The angle at each sampling time is read out as zero bias; Indicates the first k The scaling factor is read from the angle at each sampling time. express The transpose of the matrix; Or the augmented state vector is: ;in, Indicates the first k The augmented state vector at each sampling time; Indicates the first k The torsional angle at each sampling moment; Indicates the first k The angular velocity of the torsion at each sampling moment; Indicates the first k The equivalent moment of inertia at each sampling time; Indicates the first k The equivalent damping coefficient at each sampling time; This represents the torsional stiffness coefficient at the k-th sampling time. Indicates the first k The balance angle at each sampling time; Indicates the first k Input torque coefficient at each sampling time; This indicates that the angle readout at the k-th sampling time is zero bias; express The transpose of the matrix; The system parameters in the augmented state vector change with time, and the state transition is expressed as follows: ;in, Indicates the first k The system parameters to be estimated at +1 sampling time; Indicates the first k The system parameters to be estimated at each sampling time, Indicates the first k The parameter process noise at each sampling time; by setting The covariance control algorithm tracks parameter drift speed; The observation model is: ;in, Indicates the first k Angle sensor measurement values ​​at each sampling time; Indicates the first k The scaling factor is read from the angle at each sampling time. Indicates the first k The torsional angle at each sampling moment; Indicates the first k The angle at each sampling time is read out as zero bias; This represents the angle measurement noise at the k-th sampling time. S3. Perform extended Kalman filter recursive estimation; use the extended Kalman filter algorithm to predict and update the augmented state vector, and recursively output the estimated values ​​of the torsional motion state, system parameters, and estimated covariance. S4. Perform anomaly detection and output results: Determine whether the measured value is abnormal based on the normalized innovation square. If the measured value is abnormal, reduce the weight of the current observation or skip the observation update at the current time. If the measured value is normal, continue the recursive estimation and output the estimated value of the torsional motion state, the estimated value of the system parameters, and the estimated covariance.

2. The method for jointly estimating the torsional pendulum motion state and system parameters according to claim 1, characterized in that, The torsion pendulum measurement system in S1 includes: a torsion pendulum mechanical unit, an excitation or loading unit, a readout unit, a data acquisition unit, and a data processing unit; the torsion pendulum mechanical unit includes a torsion wire, a pendulum body, and a support structure; the excitation or loading unit is a device for applying a known or controllable input torque to the pendulum body; the readout unit is a sensor for measuring the torsion angle, displacement, angular velocity, or torque of the pendulum body; and the data processing unit is used to execute the extended Kalman filter algorithm. Temperature, air pressure, and vibration environment monitoring are enabled during torsion measurement. For input signals, sensor signals, and environmental signals with different sampling rates, timestamps are used for synchronization.

3. The method for jointly estimating the torsional oscillation state and system parameters according to claim 1, characterized in that, The dynamic model of the torsional pendulum system in S2 is expressed by the following equation: ;in, I This represents the equivalent moment of inertia of the torsional yaw system. c Indicates the equivalent damping coefficient; Indicates the linear torsional stiffness coefficient; Indicates the angle of torsion; Indicates the angular velocity of the torsional yaw; Indicates angular acceleration; Indicates the equilibrium angle; Indicates the input signal; Indicates the input torque coefficient; This indicates an unmodeled disturbance torque; The dynamic model of the torsional pendulum system is expressed in normalized form as follows: ; ;in, Indicates the angular velocity of the torsional yaw; Indicates angular acceleration; Represents the equivalent damping parameter. ; Represents the equivalent stiffness parameter. ; Indicates the normalized input parameters. .

4. The method for jointly estimating the torsional pendulum motion state and system parameters according to claim 1, characterized in that, In S3, the extended Kalman filter recursive estimation includes: The state prediction is based on the dynamic model of the torsional pendulum system and the state estimate from the previous moment, expressed by the following formula: in, Indicates the first k The predicted state value at each sampling time; Indicates the first k -1 state estimate at sampling time; Indicates the first k -1 sampling time of the input signal; This represents the nonlinear state transition function obtained by discretizing the dynamic model of the torsion pendulum system. Indicates the sampling period; The state transition Jacobian matrix and covariance matrix are calculated and predicted by the following formula: ; ;in, This represents the state transition Jacobian matrix at the k-th sampling time. express The transpose of the matrix; Indicates the first k -1 state estimate at sampling time; Indicates the first k State prediction covariance matrix at each sampling time; Indicates the first k -1 sampling time point state estimation covariance matrix; This represents the process noise covariance matrix at the k-th sampling time. Represents the nonlinear state transition function For the state vector The partial derivatives; The predicted observations, the observation Jacobian matrix, and the innovation quantity are calculated based on the observation model and are expressed by the following formula: ; ; ;in, This represents the predicted observation value at the k-th sampling time. Represents the observation function; This represents the predicted state value at the k-th sampling time. Represents the observation Jacobian matrix at the k-th sampling time; This represents the amount of innovation at the k-th sampling time. This represents the observation value at the k-th sampling time. Represents the observation function For the state vector The partial derivatives; The innovation covariance matrix and Kalman gain are calculated using the following formula: ; ;in, This represents the innovation covariance matrix at the k-th sampling time. Represents the observation Jacobian matrix at the k-th sampling time; express The transpose of the matrix; Indicates the first k State prediction covariance matrix at each sampling time; This represents the measurement noise covariance matrix at the k-th sampling time. This represents the Kalman gain at the k-th sampling time. express The inverse matrix; The updated augmented state vector and covariance matrix are expressed by the following equation: ; ;in, Indicates the first k State estimates at each sampling time; Indicates the first k The predicted state value at each sampling time; This represents the Kalman gain at the k-th sampling time. express The transpose of the matrix; This represents the amount of innovation at the k-th sampling time. This represents the estimation of the covariance matrix; Represents the identity matrix; Represents the observation Jacobian matrix at the k-th sampling time; express The transpose of the matrix; Indicates the first k State prediction covariance matrix at each sampling time; Let represent the measurement noise covariance matrix at the k-th sampling time.

5. The method for jointly estimating the torsional oscillation state and system parameters according to claim 1, characterized in that, The normalized squared innovation in S4 is calculated using the following formula: ;in, Represents the normalized squared innovation at the k-th sampling time; express The transpose of the matrix; express The inverse matrix; where, This represents the innovation covariance matrix at the k-th sampling time. This represents the amount of innovation at the k-th sampling time. when If the value exceeds a preset threshold, the current observation is determined to be an abnormal measurement value, and the corresponding measurement noise covariance is increased, the weight of the current observation is reduced, or the observation update at the current time is skipped. When the changes in motion state and system parameters in the augmented state vector are less than a set threshold within M consecutive sampling windows, and the average value of innovation is close to zero and the magnitude of innovation covariance matrix is ​​within a set range, the estimation of the augmented state vector is determined to have entered a stable phase; where M is the sliding window length.

6. The method for jointly estimating the torsional oscillation state and system parameters according to claim 2, characterized in that, The readout unit includes a multi-channel capacitive displacement sensor, which includes at least two channels arranged along the X direction: the voltage signal collected in real time by the multi-channel capacitive displacement sensor is converted into a torsional angle observation, and the torsional angle observation is input into the extended Kalman filter for recursive estimation point by point; For the i A capacitive displacement sensor converts voltage into displacement, expressed by the following formula. ;in, This represents the displacement measured by the i-th capacitive displacement sensor; This represents the voltage measurement value of the i-th capacitive displacement sensor; This indicates the zero-position bias of the i-th capacitive displacement sensor; This represents the proportional coefficient of the i-th capacitive displacement sensor.

7. The method for jointly estimating the torsional oscillation state and system parameters according to claim 1, characterized in that, The multi-channel capacitive displacement sensor includes two channels arranged along the X-axis; for the two capacitive displacement sensors in the X-axis, the yaw angle around the z-axis is calculated according to the following relationship: ;in, Indicates the yaw angle about the z-axis; , These represent the displacements measured by the two capacitive displacement sensors, respectively. This indicates the spacing parameter between the two capacitive displacement sensors.

Citation Information

Patent Citations

  • Measuring method of inverted torsion internal friction instrument

    CN102478549A

  • Automatic periodic measuring device of three-wire torsional pendulum vibration

    CN102809473A

  • Method for building neural network model, method for determining torsional pendulum drill string parameter and method for directional drilling

    CN110500034A

  • Sinusoidal torque device system parameter online testing method and system

    CN111537121A

  • Rigid body rotational inertia determination method based on torsional pendulum method

    CN113375863A